Datasets:
Create diffusiondb-pixelart.py
Browse files- diffusiondb-pixelart.py +408 -0
diffusiondb-pixelart.py
ADDED
|
@@ -0,0 +1,408 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Original Copyright 2022 Jay Wang, Evan Montoya, David Munechika, Alex Yang, Ben Hoover, Polo Chau
|
| 2 |
+
# MIT License
|
| 3 |
+
"""Loading script for DiffusionDB."""
|
| 4 |
+
|
| 5 |
+
import re
|
| 6 |
+
import numpy as np
|
| 7 |
+
import pandas as pd
|
| 8 |
+
|
| 9 |
+
from json import load, dump
|
| 10 |
+
from os.path import join, basename
|
| 11 |
+
from huggingface_hub import hf_hub_url
|
| 12 |
+
|
| 13 |
+
import datasets
|
| 14 |
+
|
| 15 |
+
# Find for instance the citation on arxiv or on the dataset repo/website
|
| 16 |
+
_CITATION = """\
|
| 17 |
+
@article{wangDiffusionDBLargescalePrompt2022,
|
| 18 |
+
title = {{{DiffusionDB}}: {{A}} Large-Scale Prompt Gallery Dataset for Text-to-Image Generative Models},
|
| 19 |
+
author = {Wang, Zijie J. and Montoya, Evan and Munechika, David and Yang, Haoyang and Hoover, Benjamin and Chau, Duen Horng},
|
| 20 |
+
year = {2022},
|
| 21 |
+
journal = {arXiv:2210.14896 [cs]},
|
| 22 |
+
url = {https://arxiv.org/abs/2210.14896}
|
| 23 |
+
}
|
| 24 |
+
"""
|
| 25 |
+
|
| 26 |
+
# You can copy an official description
|
| 27 |
+
_DESCRIPTION = """
|
| 28 |
+
DiffusionDB is the first large-scale text-to-image prompt dataset. It contains 2
|
| 29 |
+
million images generated by Stable Diffusion using prompts and hyperparameters
|
| 30 |
+
specified by real users. The unprecedented scale and diversity of this
|
| 31 |
+
human-actuated dataset provide exciting research opportunities in understanding
|
| 32 |
+
the interplay between prompts and generative models, detecting deepfakes, and
|
| 33 |
+
designing human-AI interaction tools to help users more easily use these models.
|
| 34 |
+
"""
|
| 35 |
+
|
| 36 |
+
_HOMEPAGE = "https://poloclub.github.io/diffusiondb"
|
| 37 |
+
_LICENSE = "CC0 1.0"
|
| 38 |
+
_VERSION = datasets.Version("0.9.1")
|
| 39 |
+
|
| 40 |
+
# Programmatically generate the URLs for different parts
|
| 41 |
+
# hf_hub_url() provides a more flexible way to resolve the file URLs
|
| 42 |
+
# https://huggingface.co/datasets/poloclub/diffusiondb/resolve/main/images/part-000001.zip
|
| 43 |
+
_URLS = {}
|
| 44 |
+
_URLS_LARGE = {}
|
| 45 |
+
_PART_IDS = range(1, 2001)
|
| 46 |
+
_PART_IDS_LARGE = range(1, 14001)
|
| 47 |
+
|
| 48 |
+
for i in _PART_IDS:
|
| 49 |
+
_URLS[i] = hf_hub_url(
|
| 50 |
+
"poloclub/diffusiondb",
|
| 51 |
+
filename=f"images/part-{i:06}.zip",
|
| 52 |
+
repo_type="dataset",
|
| 53 |
+
)
|
| 54 |
+
|
| 55 |
+
for i in _PART_IDS_LARGE:
|
| 56 |
+
if i < 10001:
|
| 57 |
+
_URLS_LARGE[i] = hf_hub_url(
|
| 58 |
+
"poloclub/diffusiondb",
|
| 59 |
+
filename=f"diffusiondb-large-part-1/part-{i:06}.zip",
|
| 60 |
+
repo_type="dataset",
|
| 61 |
+
)
|
| 62 |
+
else:
|
| 63 |
+
_URLS_LARGE[i] = hf_hub_url(
|
| 64 |
+
"poloclub/diffusiondb",
|
| 65 |
+
filename=f"diffusiondb-large-part-2/part-{i:06}.zip",
|
| 66 |
+
repo_type="dataset",
|
| 67 |
+
)
|
| 68 |
+
|
| 69 |
+
# Add the metadata parquet URL as well
|
| 70 |
+
_URLS["metadata"] = hf_hub_url(
|
| 71 |
+
"poloclub/diffusiondb", filename="metadata.parquet", repo_type="dataset"
|
| 72 |
+
)
|
| 73 |
+
|
| 74 |
+
_URLS_LARGE["metadata"] = hf_hub_url(
|
| 75 |
+
"poloclub/diffusiondb",
|
| 76 |
+
filename="metadata-large.parquet",
|
| 77 |
+
repo_type="dataset",
|
| 78 |
+
)
|
| 79 |
+
|
| 80 |
+
_SAMPLER_DICT = {
|
| 81 |
+
1: "ddim",
|
| 82 |
+
2: "plms",
|
| 83 |
+
3: "k_euler",
|
| 84 |
+
4: "k_euler_ancestral",
|
| 85 |
+
5: "ddik_heunm",
|
| 86 |
+
6: "k_dpm_2",
|
| 87 |
+
7: "k_dpm_2_ancestral",
|
| 88 |
+
8: "k_lms",
|
| 89 |
+
9: "others",
|
| 90 |
+
}
|
| 91 |
+
|
| 92 |
+
|
| 93 |
+
class DiffusionDBConfig(datasets.BuilderConfig):
|
| 94 |
+
"""BuilderConfig for DiffusionDB."""
|
| 95 |
+
|
| 96 |
+
def __init__(self, part_ids, is_large, **kwargs):
|
| 97 |
+
"""BuilderConfig for DiffusionDB.
|
| 98 |
+
Args:
|
| 99 |
+
part_ids([int]): A list of part_ids.
|
| 100 |
+
is_large(bool): If downloading data from DiffusionDB Large (14 million)
|
| 101 |
+
**kwargs: keyword arguments forwarded to super.
|
| 102 |
+
"""
|
| 103 |
+
super(DiffusionDBConfig, self).__init__(version=_VERSION, **kwargs)
|
| 104 |
+
self.part_ids = part_ids
|
| 105 |
+
self.is_large = is_large
|
| 106 |
+
|
| 107 |
+
|
| 108 |
+
class DiffusionDB(datasets.GeneratorBasedBuilder):
|
| 109 |
+
"""A large-scale text-to-image prompt gallery dataset based on Stable Diffusion."""
|
| 110 |
+
|
| 111 |
+
BUILDER_CONFIGS = []
|
| 112 |
+
|
| 113 |
+
# Programmatically generate configuration options (HF requires to use a string
|
| 114 |
+
# as the config key)
|
| 115 |
+
for num_k in [1, 5, 10, 50, 100, 500, 1000]:
|
| 116 |
+
for sampling in ["first", "random"]:
|
| 117 |
+
for is_large in [False, True]:
|
| 118 |
+
num_k_str = f"{num_k}k" if num_k < 1000 else f"{num_k // 1000}m"
|
| 119 |
+
subset_str = "large_" if is_large else "2m_"
|
| 120 |
+
|
| 121 |
+
if sampling == "random":
|
| 122 |
+
# Name the config
|
| 123 |
+
cur_name = subset_str + "random_" + num_k_str
|
| 124 |
+
|
| 125 |
+
# Add a short description for each config
|
| 126 |
+
cur_description = (
|
| 127 |
+
f"Random {num_k_str} images with their prompts and parameters"
|
| 128 |
+
)
|
| 129 |
+
|
| 130 |
+
# Sample part_ids
|
| 131 |
+
total_part_ids = _PART_IDS_LARGE if is_large else _PART_IDS
|
| 132 |
+
part_ids = np.random.choice(
|
| 133 |
+
total_part_ids, num_k, replace=False
|
| 134 |
+
).tolist()
|
| 135 |
+
else:
|
| 136 |
+
# Name the config
|
| 137 |
+
cur_name = subset_str + "first_" + num_k_str
|
| 138 |
+
|
| 139 |
+
# Add a short description for each config
|
| 140 |
+
cur_description = f"The first {num_k_str} images in this dataset with their prompts and parameters"
|
| 141 |
+
|
| 142 |
+
# Sample part_ids
|
| 143 |
+
total_part_ids = _PART_IDS_LARGE if is_large else _PART_IDS
|
| 144 |
+
part_ids = total_part_ids[1 : num_k + 1]
|
| 145 |
+
|
| 146 |
+
# Create configs
|
| 147 |
+
BUILDER_CONFIGS.append(
|
| 148 |
+
DiffusionDBConfig(
|
| 149 |
+
name=cur_name,
|
| 150 |
+
part_ids=part_ids,
|
| 151 |
+
is_large=is_large,
|
| 152 |
+
description=cur_description,
|
| 153 |
+
),
|
| 154 |
+
)
|
| 155 |
+
|
| 156 |
+
# Add few more options for Large only
|
| 157 |
+
for num_k in [5000, 10000]:
|
| 158 |
+
for sampling in ["first", "random"]:
|
| 159 |
+
num_k_str = f"{num_k // 1000}m"
|
| 160 |
+
subset_str = "large_"
|
| 161 |
+
|
| 162 |
+
if sampling == "random":
|
| 163 |
+
# Name the config
|
| 164 |
+
cur_name = subset_str + "random_" + num_k_str
|
| 165 |
+
|
| 166 |
+
# Add a short description for each config
|
| 167 |
+
cur_description = (
|
| 168 |
+
f"Random {num_k_str} images with their prompts and parameters"
|
| 169 |
+
)
|
| 170 |
+
|
| 171 |
+
# Sample part_ids
|
| 172 |
+
total_part_ids = _PART_IDS_LARGE
|
| 173 |
+
part_ids = np.random.choice(
|
| 174 |
+
total_part_ids, num_k, replace=False
|
| 175 |
+
).tolist()
|
| 176 |
+
else:
|
| 177 |
+
# Name the config
|
| 178 |
+
cur_name = subset_str + "first_" + num_k_str
|
| 179 |
+
|
| 180 |
+
# Add a short description for each config
|
| 181 |
+
cur_description = f"The first {num_k_str} images in this dataset with their prompts and parameters"
|
| 182 |
+
|
| 183 |
+
# Sample part_ids
|
| 184 |
+
total_part_ids = _PART_IDS_LARGE
|
| 185 |
+
part_ids = total_part_ids[1 : num_k + 1]
|
| 186 |
+
|
| 187 |
+
# Create configs
|
| 188 |
+
BUILDER_CONFIGS.append(
|
| 189 |
+
DiffusionDBConfig(
|
| 190 |
+
name=cur_name,
|
| 191 |
+
part_ids=part_ids,
|
| 192 |
+
is_large=True,
|
| 193 |
+
description=cur_description,
|
| 194 |
+
),
|
| 195 |
+
)
|
| 196 |
+
|
| 197 |
+
# Need to manually add all (2m) and all (large)
|
| 198 |
+
BUILDER_CONFIGS.append(
|
| 199 |
+
DiffusionDBConfig(
|
| 200 |
+
name="2m_all",
|
| 201 |
+
part_ids=_PART_IDS,
|
| 202 |
+
is_large=False,
|
| 203 |
+
description="All images with their prompts and parameters",
|
| 204 |
+
),
|
| 205 |
+
)
|
| 206 |
+
|
| 207 |
+
BUILDER_CONFIGS.append(
|
| 208 |
+
DiffusionDBConfig(
|
| 209 |
+
name="large_all",
|
| 210 |
+
part_ids=_PART_IDS_LARGE,
|
| 211 |
+
is_large=True,
|
| 212 |
+
description="All images with their prompts and parameters",
|
| 213 |
+
),
|
| 214 |
+
)
|
| 215 |
+
|
| 216 |
+
# We also prove a text-only option, which loads the meatadata parquet file
|
| 217 |
+
BUILDER_CONFIGS.append(
|
| 218 |
+
DiffusionDBConfig(
|
| 219 |
+
name="2m_text_only",
|
| 220 |
+
part_ids=[],
|
| 221 |
+
is_large=False,
|
| 222 |
+
description="Only include all prompts and parameters (no image)",
|
| 223 |
+
),
|
| 224 |
+
)
|
| 225 |
+
|
| 226 |
+
BUILDER_CONFIGS.append(
|
| 227 |
+
DiffusionDBConfig(
|
| 228 |
+
name="large_text_only",
|
| 229 |
+
part_ids=[],
|
| 230 |
+
is_large=True,
|
| 231 |
+
description="Only include all prompts and parameters (no image)",
|
| 232 |
+
),
|
| 233 |
+
)
|
| 234 |
+
|
| 235 |
+
# Default to only load 1k random images
|
| 236 |
+
DEFAULT_CONFIG_NAME = "2m_random_1k"
|
| 237 |
+
|
| 238 |
+
def _info(self):
|
| 239 |
+
"""Specify the information of DiffusionDB."""
|
| 240 |
+
|
| 241 |
+
if "text_only" in self.config.name:
|
| 242 |
+
features = datasets.Features(
|
| 243 |
+
{
|
| 244 |
+
"image_name": datasets.Value("string"),
|
| 245 |
+
"prompt": datasets.Value("string"),
|
| 246 |
+
"part_id": datasets.Value("uint16"),
|
| 247 |
+
"seed": datasets.Value("uint32"),
|
| 248 |
+
"step": datasets.Value("uint16"),
|
| 249 |
+
"cfg": datasets.Value("float32"),
|
| 250 |
+
"sampler": datasets.Value("string"),
|
| 251 |
+
"width": datasets.Value("uint16"),
|
| 252 |
+
"height": datasets.Value("uint16"),
|
| 253 |
+
"user_name": datasets.Value("string"),
|
| 254 |
+
"timestamp": datasets.Value("timestamp[us, tz=UTC]"),
|
| 255 |
+
"image_nsfw": datasets.Value("float32"),
|
| 256 |
+
"prompt_nsfw": datasets.Value("float32"),
|
| 257 |
+
},
|
| 258 |
+
)
|
| 259 |
+
|
| 260 |
+
else:
|
| 261 |
+
features = datasets.Features(
|
| 262 |
+
{
|
| 263 |
+
"image": datasets.Image(),
|
| 264 |
+
"prompt": datasets.Value("string"),
|
| 265 |
+
"seed": datasets.Value("uint32"),
|
| 266 |
+
"step": datasets.Value("uint16"),
|
| 267 |
+
"cfg": datasets.Value("float32"),
|
| 268 |
+
"sampler": datasets.Value("string"),
|
| 269 |
+
"width": datasets.Value("uint16"),
|
| 270 |
+
"height": datasets.Value("uint16"),
|
| 271 |
+
"user_name": datasets.Value("string"),
|
| 272 |
+
"timestamp": datasets.Value("timestamp[us, tz=UTC]"),
|
| 273 |
+
"image_nsfw": datasets.Value("float32"),
|
| 274 |
+
"prompt_nsfw": datasets.Value("float32"),
|
| 275 |
+
},
|
| 276 |
+
)
|
| 277 |
+
|
| 278 |
+
return datasets.DatasetInfo(
|
| 279 |
+
description=_DESCRIPTION,
|
| 280 |
+
features=features,
|
| 281 |
+
supervised_keys=None,
|
| 282 |
+
homepage=_HOMEPAGE,
|
| 283 |
+
license=_LICENSE,
|
| 284 |
+
citation=_CITATION,
|
| 285 |
+
)
|
| 286 |
+
|
| 287 |
+
def _split_generators(self, dl_manager):
|
| 288 |
+
# If several configurations are possible (listed in BUILDER_CONFIGS),
|
| 289 |
+
# the configuration selected by the user is in self.config.name
|
| 290 |
+
|
| 291 |
+
# dl_manager is a datasets.download.DownloadManager that can be used to
|
| 292 |
+
# download and extract URLS It can accept any type or nested list/dict
|
| 293 |
+
# and will give back the same structure with the url replaced with path
|
| 294 |
+
# to local files. By default the archives will be extracted and a path
|
| 295 |
+
# to a cached folder where they are extracted is returned instead of the
|
| 296 |
+
# archive
|
| 297 |
+
|
| 298 |
+
# Download and extract zip files of all sampled part_ids
|
| 299 |
+
data_dirs = []
|
| 300 |
+
json_paths = []
|
| 301 |
+
|
| 302 |
+
# Resolve the urls
|
| 303 |
+
if self.config.is_large:
|
| 304 |
+
urls = _URLS_LARGE
|
| 305 |
+
else:
|
| 306 |
+
urls = _URLS
|
| 307 |
+
|
| 308 |
+
for cur_part_id in self.config.part_ids:
|
| 309 |
+
cur_url = urls[cur_part_id]
|
| 310 |
+
data_dir = dl_manager.download_and_extract(cur_url)
|
| 311 |
+
|
| 312 |
+
data_dirs.append(data_dir)
|
| 313 |
+
json_paths.append(join(data_dir, f"part-{cur_part_id:06}.json"))
|
| 314 |
+
|
| 315 |
+
# Also download the metadata table
|
| 316 |
+
metadata_path = dl_manager.download(urls["metadata"])
|
| 317 |
+
|
| 318 |
+
return [
|
| 319 |
+
datasets.SplitGenerator(
|
| 320 |
+
name=datasets.Split.TRAIN,
|
| 321 |
+
# These kwargs will be passed to _generate_examples
|
| 322 |
+
gen_kwargs={
|
| 323 |
+
"data_dirs": data_dirs,
|
| 324 |
+
"json_paths": json_paths,
|
| 325 |
+
"metadata_path": metadata_path,
|
| 326 |
+
},
|
| 327 |
+
),
|
| 328 |
+
]
|
| 329 |
+
|
| 330 |
+
def _generate_examples(self, data_dirs, json_paths, metadata_path):
|
| 331 |
+
# This method handles input defined in _split_generators to yield
|
| 332 |
+
# (key, example) tuples from the dataset.
|
| 333 |
+
# The `key` is for legacy reasons (tfds) and is not important in itself,
|
| 334 |
+
# but must be unique for each example.
|
| 335 |
+
|
| 336 |
+
# Load the metadata parquet file if the config is text_only
|
| 337 |
+
if "text_only" in self.config.name:
|
| 338 |
+
metadata_df = pd.read_parquet(metadata_path)
|
| 339 |
+
for _, row in metadata_df.iterrows():
|
| 340 |
+
yield row["image_name"], {
|
| 341 |
+
"image_name": row["image_name"],
|
| 342 |
+
"prompt": row["prompt"],
|
| 343 |
+
"part_id": row["part_id"],
|
| 344 |
+
"seed": row["seed"],
|
| 345 |
+
"step": row["step"],
|
| 346 |
+
"cfg": row["cfg"],
|
| 347 |
+
"sampler": _SAMPLER_DICT[int(row["sampler"])],
|
| 348 |
+
"width": row["width"],
|
| 349 |
+
"height": row["height"],
|
| 350 |
+
"user_name": row["user_name"],
|
| 351 |
+
"timestamp": None
|
| 352 |
+
if pd.isnull(row["timestamp"])
|
| 353 |
+
else row["timestamp"],
|
| 354 |
+
"image_nsfw": row["image_nsfw"],
|
| 355 |
+
"prompt_nsfw": row["prompt_nsfw"],
|
| 356 |
+
}
|
| 357 |
+
|
| 358 |
+
else:
|
| 359 |
+
num_data_dirs = len(data_dirs)
|
| 360 |
+
assert num_data_dirs == len(json_paths)
|
| 361 |
+
|
| 362 |
+
# Read the metadata table (only rows with the needed part_ids)
|
| 363 |
+
part_ids = []
|
| 364 |
+
for path in json_paths:
|
| 365 |
+
cur_id = int(re.sub(r"part-(\d+)\.json", r"\1", basename(path)))
|
| 366 |
+
part_ids.append(cur_id)
|
| 367 |
+
|
| 368 |
+
# We have to use pandas here to make the dataset preview work (it
|
| 369 |
+
# uses streaming mode)
|
| 370 |
+
metadata_table = pd.read_parquet(
|
| 371 |
+
metadata_path,
|
| 372 |
+
filters=[("part_id", "in", part_ids)],
|
| 373 |
+
)
|
| 374 |
+
|
| 375 |
+
# Iterate through all extracted zip folders for images
|
| 376 |
+
for k in range(num_data_dirs):
|
| 377 |
+
cur_data_dir = data_dirs[k]
|
| 378 |
+
cur_json_path = json_paths[k]
|
| 379 |
+
|
| 380 |
+
json_data = load(open(cur_json_path, "r", encoding="utf8"))
|
| 381 |
+
|
| 382 |
+
for img_name in json_data:
|
| 383 |
+
img_params = json_data[img_name]
|
| 384 |
+
img_path = join(cur_data_dir, img_name)
|
| 385 |
+
|
| 386 |
+
# Query the metadata
|
| 387 |
+
query_result = metadata_table.query(f'`image_name` == "{img_name}"')
|
| 388 |
+
|
| 389 |
+
# Yields examples as (key, example) tuples
|
| 390 |
+
yield img_name, {
|
| 391 |
+
"image": {
|
| 392 |
+
"path": img_path,
|
| 393 |
+
"bytes": open(img_path, "rb").read(),
|
| 394 |
+
},
|
| 395 |
+
"prompt": img_params["p"],
|
| 396 |
+
"seed": int(img_params["se"]),
|
| 397 |
+
"step": int(img_params["st"]),
|
| 398 |
+
"cfg": float(img_params["c"]),
|
| 399 |
+
"sampler": img_params["sa"],
|
| 400 |
+
"width": query_result["width"].to_list()[0],
|
| 401 |
+
"height": query_result["height"].to_list()[0],
|
| 402 |
+
"user_name": query_result["user_name"].to_list()[0],
|
| 403 |
+
"timestamp": None
|
| 404 |
+
if pd.isnull(query_result["timestamp"].to_list()[0])
|
| 405 |
+
else query_result["timestamp"].to_list()[0],
|
| 406 |
+
"image_nsfw": query_result["image_nsfw"].to_list()[0],
|
| 407 |
+
"prompt_nsfw": query_result["prompt_nsfw"].to_list()[0],
|
| 408 |
+
}
|